DoorDash scaled a merchant-facing analytics product from a new use case to thousands of users within a few weeks, according to Harsha Reddy, a senior engineering leader who heads Data End-User Platforms in the company's data platform organization. He discussed the path from prototype to production in a fireside chat with Sonny Rivera of ThoughtSpot.
The core challenge was balancing central governance with user velocity, made harder by acquisitions. Deliveroo in the UK and Wolt in Finland and much of the EU each brought their own semantic layers, often built in Looker, on top of DoorDash's in-house semantic tooling. Consolidating everything into one governed layer without slowing teams down became the real problem.
DoorDash's answer splits the semantic layer from the consumption layer. Data lands in a data lake exposed as Iceberg tables and in Snowflake, dbt defines the semantic and metric layer on top of both, and ThoughtSpot connects directly to dbt's metric definitions instead of keeping a parallel set. Reddy said that beats an all-in-one tool like Looker, where everything is encoded in one place and is not much use to anyone else.
Internally, finance and revenue reporting, where the margin for error is zero, now relies on the governed layer plus ThoughtSpot's compliance and audit reporting, replacing manual work done to satisfy GDPR, HIPAA and other requirements. Externally, DoorDash embeds ThoughtSpot fully white-labeled in merchant tools, with role-level access control so one merchant never sees another's data.
Reddy's advice: semantic modeling is hard, but when generating an answer is cheap and trusting it is what matters, the effort pays off. Treat the semantic layer as code, version-controlled and reviewed, and make the governed path at least as easy as an ad hoc query, or people will route around it. DoorDash has also moved from domain-specific agents behind a router to a single global agent grounded in the same dbt-defined semantic layer, covering roughly 70 to 80 percent of its data domains.
Source: dbt Labs Blog